Urban air pollution by odor sources: short time prediction Nicola Pettarinb,c , Marina Campolo1 anda , Alfredo Soldatib,c a
Dept. Chemistry, Physics & Environment, University of Udine, 33100 Udine, Italy b Dept. Management, Electric & Mechanical Engineering, University of Udine, 33100 Udine, Italy c LOD srl, 33100 Udine, Italy
Abstract A numerical approach is proposed to predict the short time dispersion of odors in the urban environment. The model is based on (i) a three dimensional computational domain describing the urban topography at fine spatial scale (one meter) and on (ii) highly time resolved (one minute frequency) meteorological data used as inflow conditions. The time dependent, three dimensional wind velocity field is reconstructed in the Eulerian framework using a fast response finite volume solver of Navier-Stokes equations. Odor dispersion is calculated using a Lagrangian approach. An application of the model to the historic city of Verona (Italy) is presented. Results confirm that this type of odor dispersion simulations can be used (i) to assess the impact of odor emissions in urban areas and (ii) to evaluate the potential mitigation produced by odor abatement systems. Keywords: Dispersion modelling, short averaging time, odor pollution, 1
Corresponding author. Address via Cotonificio 108, 33100 Udine (UD), Italy; e-mail
[email protected]; phone: +39 432 558822; facs: +39 432 558803
Accepted for publication in Atmospheric Environment
September 11, 2015
QUIC
1
1. Introduction
2
Exposure to unpleasant odors is one of the most frequent causes of air
3
quality complaints in both industrial and urban areas. The chemical com-
4
pounds responsible for odor generation are volatile species (Olafsdottir and
5
Gardarsson, 2013): once emitted from a source, their transport, dispersion
6
and fate in the environment is controlled by the complex interaction among
7
strength of emission (Campolo et al., 2005), meteorological conditions and
8
site topography. Odors become perceptible whenever the instantaneous and
9
local concentration of these chemicals transpasses very low concentration val-
10
ues corresponding to the odor detection threshold. This may occur nearby a
11
source but also some distance away from it.
12
Odor perception is synchronous with breathing and involuntary, but the
13
subsequent reaction to a given odor stimulus is to some degree subjective:
14
it depends on odor intensity and offensiveness, duration and frequency of
15
exposure but also on pleasantness/unpleasantness of the sensation evoked
16
by the odor (Blanes-Vidal et al., 2012). Annoyance may be produced from
17
acute exposure to few, high odor intensity events or to chronic exposure to
18
repeated, low odor intensity events (Griffiths, 2014). Whichever the expo-
19
sure mode, odors generating a negative appraisal induce changes in people
20
behavior and may trigger a stress-mediated response which may develop into
21
a public health concern. Bad smells which occasionally cause annoyance,
22
are proactively reported to the Health Services: when the source of the odor
23
can be clearly identified and associated to a specific emission either by the 2
24
analysis of resident nuisance odor reports (Nicolas et al., 2011), by the use
25
of chemical sensors (Sohn et al., 2009; Seo et al., 2011) or by other sensory
26
methods (Brattoli et al., 2011, Capelli et al., 2011), corrective actions can be
27
devised as needed to contain/reduce the odor impact.
28
Odor impact in urban areas can be very difficult to assess and control
29
due to the inherent complexity of the urban environment, the large num-
30
ber of potential sources and the local small scale variability of the dispers-
31
ing wind. Odor nuisance is most frequently associated with discontinuous
32
emissions generated by restaurants, fast food and bar which may occur for
33
short/prolonged times (from a few seconds to minutes), occasionally or on
34
a repetitive basis depending on the actual operating hours of the facility.
35
The odor impact potentially arising from these commercial activities should
36
be taken into account when planning new installations: best practices for
37
design and operation of commercial kitchen ventilation systems have been
38
developed (see DEFRA, 2005) and yet more accurate modelling tools could
39
be profitably used for odour pollution assessment, prevention and mitiga-
40
tion. Odor emissions in a high populated urban area could be confidently
41
authorized if the potential impact of each source could be estimated a priori
42
by modelling; moreover, the precise evaluation of the odour impact of an ex-
43
isting source might be required for the detailed analysis of resident nuisance
44
odor reports in support of litigations for odor impact problems.
45
Odor impact assessment based on chemical sensors would require the ac-
46
quisition of highly time resolved, compound specific, qualified low-concentration
47
data which are very difficult to obtain experimentally. Furthermore, most
48
odors are generated by mixtures of compounds and the relationship between
3
49
species concentration and odor nuisance is not straightforward. A more prac-
50
tical and effective approach may be the numerical prediction of odor disper-
51
sion.
52
Numerical models have been successfully used to predict odor dispersion
53
and to assess odor impact in industrial areas (see Nicell, 2009, Sironi et al.,
54
2010). The common approach is to model the odor as a passive chemical,
55
equivalent to the mixture of chemicals present, whose concentration is con-
56
veniently represented by the number of odor units, a multiple of the mixture
57
detection threshold. Most of the models in use has been adapted from ear-
58
lier studies on air pollution: steady state Gaussian plume models (Latos et
59
al., 2011), fluctuating plume models (Mussio et al., 2001; Dourado et al.,
60
2014) and Lagrangian stochastic dispersion models (Franzese, 2003) have
61
been used. The main challenge when using these models to predict odor
62
dispersion is related with the different time and space resolution at which
63
the prediction is required. The time scale of few seconds (corresponding to a
64
single human breath) required to evaluate odor impact is much smaller than
65
the hourly time scale typically used to evaluate the dispersion of pollutant
66
species. If a hourly time scale is maintained for odor dispersion modelling,
67
the peak odor concentration at the time scale relevant for odor impact as-
68
sessment should be estimated using a peak to mean ratio, which can be
69
either assumed to be constant (Sironi et al., 2010) or calculated based on
70
wind speed, atmospheric stability, distance from and geometry of the source
71
(Piringer et al., 2012; Schauberger et al., 2012).
72
Very different regulation limits and guidelines have been used worldwide
73
to fix benchmark concentration for odors: Nicell (2009) reports values of off-
4
74
site odor limits ranging from 0.5 to 50 odor units, averaging time ranging
75
from 1 s to 1 hr and compliance frequency ranging from 98% to 100%. In
76
Australia odor criteria (based on 3 minute average and 99.9% frequency) are
77
population density dependent (see EPA 373/07). The large variability in
78
odor exposure criteria indicates that there is still little consensus on what
79
odor concentration and/or averaging time represent the most effective and
80
fair odor limits for off-site impact. Recently, odor criteria have been clas-
81
sified into two groups (Sommer-Quabach et al., 2014): those based on low
82
odor concentration threshold and high exceedance frequency, relevant to as-
83
sess chronic exposure, and those based on high concentration threshold and
84
low exceedance frequency, relevant to assess acute exposure. At now, the rec-
85
ommended approach for odor regulation in Europe belongs to the first type
86
(chronic exposure oriented) and consists in predicting by numerical models
87
the hourly mean of odor concentration for at least one year period (up to 3 or
88
5 years) and to check odor exposure considering the 98th percentile of those
89
data (see Environment Agency, 2011). The choice of the 98th percentile
90
is supported by the strong correlation found with annoyance measured by
91
community surveys (see Pullen and Vawda, 2007). Yet, different assessment
92
tools and regulatory responses may be required to effectively manage acute
93
exposure scenario (Griffiths, 2014).
94
A possibility is to use a smaller time scale for the odor dispersion mod-
95
elling by which the peaks in odor concentration which result in annoyance
96
for the population can be directly captured: Drew et al. (2007) demon-
97
strated that dispersion modelling based on short averaging time was more
98
successful than the current regulatory method at capturing odor peak con-
5
99
centrations from a landfill site. Peak odor intensity is often associated with
100
relatively weak meteorological dynamics (light winds) for which short term
101
and short range effects may be important: wind directions can be highly vari-
102
able (Huiling-cui et al., 2011), turbulent motions may be of the same order
103
as wind speed and the shear production term may dominate in the turbulent
104
kinetic energy budget equation (Manor, 2014) making the turbulent trans-
105
port of species more sensitive to the presence of boundaries (complex terrain
106
and presence of buildings) and highly anisotropic (Pitton et al., 2012).
107
Eulerian-Eulerian models based on Reynolds Averaged Navier Stokes
108
(RANS) equations and Large Eddy Simulation (LES) have proven to be
109
accurate to simulate the dispersion of chemical species (pollutants) in com-
110
plex three dimensional domains (Hanna et al., 2006). Gailis et al. (2007)
111
investigated tracer dispersion in a boundary layer sheared by a large array of
112
obstacles using a Lagrangian stochastic plume model. They found that inter-
113
nal plume fluctuations can have a greater effect on tracer dispersion than the
114
meander motion of the plume, which may be significantly damped in a rough-
115
walled boundary layer. Michioka et al. (2013) implemented a short term,
116
highly resolved (10 s) microscale large-eddy simulation (LES) model coupled
117
to a mesoscale LES model to estimate the concentration of a tracer gas in
118
an urban district considering both the influence of meteorological variability
119
and topographic effects. Their results underlined the key role of coupling
120
between mesoscale and local atmospheric dynamics in driving the dispersion
121
of tracer gas.
122
The same type of short term, fine scale models can be used to simulate
123
odor dispersion in the urban environment. Odor dispersion under steady
6
124
wind and constant emission in the presence of few buildings has been eval-
125
uated using Eulerian-Eulerian Re-Normalisation Group (RNG) k − ǫ model
126
by Maizi et al. (2010), using Large Eddy Simulation (LES) by Dourado et
127
al. (2012) and using a fluctuating plume model by Dourado et al. (2014).
128
Despite the increasing number of applications based on local, short averaging
129
time dispersion models, this modelling approach has not yet been adequately
130
validated to be confidently used for odor impact assessment (Pullen and
131
Vawda, 2007). Moreover, we are not aware of applications of odor dispersion
132
models to more complex urban environments. One of the reason is most likely
133
the high cost associated with the time-dependent, fine resolved calculations
134
needed to characterize the flow field of the carrier fluid and the transport of
135
dispersed species into a complex domain. A fine spatial grid resolution (order
136
of 1-2 meters) is required to model faithfully the complex urban domain and
137
a fine time resolution is required to model concentration fluctuations and to
138
capture the peak values responsible of the impact (see Pullen and Vawda,
139
2007). The simulation of local atmospheric dynamics highly resolved in space
140
and time may become cheaper if representative scenarios rather than full year
141
periods can be identified and considered. Moreover, cost/time of computa-
142
tion can be reduced adopting fast response models of Eulerian-Lagrangian
143
type developed and used successfully to calculate dispersion of species in
144
urban environments (Gowardhan et al., 2011).
145
In this work we propose the use of one of these models (QUIC − Quic
146
Urban & Industrial Complex model, Los Alamos Laboratories) (Gowardhan
147
et al., 2011) to evaluate the impact of odor emissions in urban environments.
148
The work is based on the assumption that the local wind field and turbulence
7
149
controlling dispersion is triggered by the urban geometry more than by the
150
microscale wind and atmospheric turbulence. Our objective is to demonstrate
151
how different can be the odor impact evaluated in the short term when the
152
dynamic interaction between wind field and complex urban topography is
153
accounted for. We will use highly resolved (one minute frequency) microscale
154
wind velocity data to reconstruct the flow field around buildings; this flow
155
field will then be used to simulate the transport of odor, to evaluate odor
156
exposure in terms of frequency of exceedance and intensity and to assess the
157
potential odor impact. To demonstrate our idea, two different meteorological
158
scenarios will be considered. Increasing the number of simulated scenarios
159
enough to cover all the meteorological conditions that may influence the
160
impact, the model could become a powerful tool to help Public Authorities
161
in their planning and control activities.
162
First, we will to demonstrate that the proposed model can be used to
163
evaluate comparatively the odor impact of a given emission source when
164
located in alternative positions inside the urban micro-environment; second,
165
we will prove that the model can be used to check if the odor impact can be
166
sufficiently abated by the installation of odor control systems. The potential
167
of the model will be demonstrated comparing the effect of untreated/treated
168
emission associated to the planned installation of fast food activities in two
169
different urban zones in the historical city of Verona (Italy).
8
170
2. Methods, site and data
171
2.1. Numerical model
172
The model proposed (QUIC) is a 3D finite volume solver of Reynolds-
173
Averaged Navier-Stokes (RANS) equations for incompressible flow. The
174
model is implemented and runs in the Matlab environment. The compu-
175
tational domain, corresponding to an urban area including a large number
176
of buildings, is defined using a structured grid in which solid/fluid cells are
177
identified using numerical coding (zero and one identify solid and fluid cells,
178
respectively). The grid is generated from Environmental Systems Research
179
Institute (ESRI) shape files using the code built-in pre-processor.
180
RANS equations are solved explicitly in time on a staggered mesh using
181
a projection method. The discretization scheme is second order accurate
182
in space and time (see Gowardhan et al., 2011 for further details). A zero
183
equation (algebraic) turbulence model is used. Free slip conditions are used
184
at the top and side boundaries of the computational domain; a prescribed,
185
time dependent velocity profile derived from an urban meteorological station
186
can be imposed at the upwind side while an outflow boundary condition is
187
imposed at the downwind side.
188
A Lagrangian particle approach is used to model odor dispersion: thou-
189
sands of ”particles” released from the emission point are tracked as they
190
are randomly advected and dispersed over the domain (Zwack et al., 2011).
191
Particles are modeled as infinitesimally small, neutrally buoyant gas parcels.
192
For the present application, a steady state emission is considered for the odor
193
plume: each particle is associated with a fraction of the odor emission rate
194
and is tracked using a small time step (0.1 seconds). Overall, about half 9
195
a million QUIC particles were released over the simulation time period (15
196
minutes).
197
Odor concentrations are determined in the Eulerian reference frame by
198
counting how many particles pass through a given computational volume
199
during the time averaging period of interest (30 seconds in our demo).
200
2.2. Urban district
201
Figure 1 (a) shows an aerial view of Verona downtown (near to the Arena).
202
Two different zones were selected for modelling odor dispersion to check
203
whether the specific localization of the source could significantly affect odor
204
impact: the first area (230×290 m wide), identified as Area 1, is characterized
205
by street canyons; the second area (495 × 250 m wide), identified as Area
206
2, faces the open square of the Arena. Figures 1 (b) and (c) show the two
207
computational models which extend 50 m above the ground.
208
The potential positions of the odor emission source in Area 1 and Area
209
2 are shown as red (light gray) dots S1 and S2. The emission height was
210
fixed as one meter above the roof level. In the local coordinates system,
211
with the grid origin at the lower left corner of each area, source positions
212
are identified by (x, y, z) triples equal to (138.5,176.5,18.5) for Area 1 and
213
(161.5,238.5,17.5) for Area 2. The blue (dark gray) circles indicate control
214
points P1 and P2 located 50 m downstream the source in the prevailing wind
215
blowing direction. The elevation of control points is 1.5 m above the ground.
216
2.3. Meteorological data
217
Data used in this work are taken from the urban station of Verona Golo-
218
sine (latitude 45o28′ 51′′ , longitude 10o 52′ 35′′ , 61 m above sea level). One10
219
minute time resolved records of wind speed and direction collected during
220
February 2012 were made available from MeteoVerona. One week of data was
221
statistically analysed. Statistics suggest that the prevailing wind blowing di-
222
rection is from Nord, North-East (N-E) and the average wind speed is about
223
0.89 m/s at the wind monitoring station (10 m elevation above the ground).
224
To demonstrate how different can be the odor impact evaluated in the short
225
term when time dependent winds interact with a complex urban topography,
226
two 15 minute long periods were extracted for modelling odor dispersion: the
227
first, event 1, is characterized by wind intensity of 3.12 ± 0.67 m/s (average
228
plus standard deviation) and wind blowing from direction 48 ± 44o degrees
229
N (average plus standard deviation); the second, event 2, is characterized
230
by wind intensity of 3.3 ± 1.2 m/s and wind blowing from direction 2 ± 20o
231
degrees N. Even if average wind intensity is similar, variability of wind in-
232
tensity is larger for event 2, whereas wind directions differ both in average
233
value and variability. The two events selected are examples of “similar” and
234
yet substantially different scenarios which need to be simulated to obtain
235
a consistent evaluation of odor impact. Considering the size of computa-
236
tional domain and average wind intensity, each 15 minute long period is long
237
enough to track the dispersion of the odor plume up to the boundaries of the
238
computational domain. More/longer periods could be routinely simulated
239
once extended meteorological data are made available.
240
Figure 2 shows the wind variation of the two selected events using a
241
polar representation (Figure 2 (a)) and time series plots of wind speed and
242
direction (Figure 2 (b) and (c)). At each time step, the direction from which
243
the wind is blowing identifies the upstream side of the computational domain;
11
244
the vertical profile of wind velocity used as inflow condition is defined by the
245
wind speed recording at the anemometer (red arrow in Figure 1 (b)) using a
246
power law.
247
2.4. Emission data
248
To characterize the strength of the emission, we considered a restaurant
249
using the same cooking methods (deep frying and stewing) of the planned
250
fast food installation. Samples used to quantify the odor emission rate were
251
collected from the chimney of the restaurant when a frying food system was
252
active. The mean cooking time for lunch (or dinner) period was 109 minutes.
253
The stack diameter was 1 m. The mean values of stack outlet velocity and
254
exhaust flow rate were 4.12 m/s and 350 Nm3 /min. The mean stack inlet
255
and outlet temperatures were 44o C and 31o C. The variability of the source
256
was checked during sampling according to EN ISO 16911:2013. We collected
257
three samples according to EN 13725:2003 using a vacuum pump to suck air
258
from the emitting stack into Nalophan bags (8 L volume); sampling required
259
about 1.5 minutes for each sample, with 10 minute stop between samples
260
to check emission variability over time; odor samples were then transferred
261
to the lab for the sensory evaluation of odors off site by a group of trained
262
panels. Mixtures of sampled air and neutral air at decreasing dilution ratio
263
were sequentially prepared by the olfactometer and smelt by the panels. The
264
test started from an odor sample which was very diluted. The dilution ratio
265
was gradually reduced up to the identification of the odor threshold, i.e. the
266
point at which the odor is only just detectable to 50% of the test panel. The
267
numerical value of the dilution ratio necessary to reach the odor threshold
268
was taken as the measure of the odor concentration at the source expressed 12
269
in European odor units per cubic meter (o.u.E /m3 , ou/m3 in brief).
270
Sampling was performed in two different working conditions, correspond-
271
ing to off/on operation for the activated carbon filter installed for odor con-
272
trol. Data collected during sampling are summarized in Table 1. Data vari-
273
ability during sampling and among samples was found to be not significant
274
and odor emission rates used to set up the model are values averaged over
275
the three samples.
276
3. Results
277
3.1. Flow field
278
The QUIC code calculates the flow field in the three dimensional do-
279
main every one minute. Figure 3 shows the comparison between the wind
280
speed/direction measured at the meteorological station (10 m height, line
281
with circles) and used as inflow condition, and those calculated in differ-
282
ent points of the computational domain: at the source (1 m above roof level,
283
empty triangles) and at the reference control point (solid triangles) for Area 1
284
(triangles pointing upward) and Area 2 (triangles pointing downward) (1.5 m
285
above ground). The effect of urban topography is to produce local differences
286
in wind intensity and direction calculated at different points.
287
Wind speed and direction calculated at the emission point (i.e. above
288
the buildings) are similar to the values recorded at the meteorological sta-
289
tion: the wind speed is a bit larger at the emission point since it is more
290
elevated than the anemometric sensor. At control points, the wind speed
291
is generally smaller than the sensor due to the different elevation above the
292
ground (1.5 m); the wind direction may be significantly different. For control 13
293
points located in a street canyon, the effect of the urban topography is to
294
smooth out the variability of wind direction. For wind event 1, the local wind
295
direction is about 50o N whichever the value recorded at the meteorological
296
station for both control points P1 and P2; for event 2, the wind direction
297
is similar at the meteorological station and point P2, whereas it is always
298
about 50o N for point P1.
299
3.2. Odor dispersion
300
Animations of the odor plume dispersing from sources S1 and S2 during
301
the two simulated wind events are available as supplementary material. The
302
position of the emission point is indicated by the black circle; isocontours
303
represent the odor concentration (in ou/m3 ) calculated in the plane 1.5 m
304
above the ground (reference height of human noses potentially smelling in
305
the area). Figures 4-5 shows snapshots (one every 240 seconds) taken from
306
the animations. The color scale for odor concentration shown in the plots
307
in limited to the sub-range [2 ÷ 12 ou/m3 ]. To relate odor concentration to
308
perceived odor intensity in the field we refer to the following scale (Sommer-
309
Quabach et al., 2014): non detectable (C < 2 ou/m3 ), acceptable (2
15 ou/m3 ). The lower and upper values of the color scale represent
312
an odor concentration threshold at which the odor is clearly detected and a
313
value at which the odor perceived is strong enough to cause annoyance.
314
Isocontours calculated during wind event 1 in Area 1 (Figure 4 upper
315
row) show the odor plume extending in different directions depending on
316
the leading wind. Yet, the urban topography determines a preferential path
317
for odor dispersion which spreads along the main street canyons near to the 14
318
source. Due to the changing wind direction, some of the odor puffs may reach
319
regions not directly exposed to the emitting source, producing diffuse odor
320
impact even at significant distances. During wind event 2, isocontours (Fig-
321
ure 4 bottom row) show odor puffs moving along three main street canyons
322
(aligned with the wind blowing directions) with odor concentration mainly
323
controlled by wind speed. Odor dispersion produced in Area 2 (open area
324
facing the Arena) for wind event 1 (Figure 5 top row) indicates that odor
325
puffs remain confined along the prevailing wind direction (from N-E to S-
326
W) despite the wind direction variability, and may penetrate into the urban
327
topography when the blowing wind direction is from S-E. For wind event 2
328
(Figure 5 bottom row) the odor plume oscillates back and forth in the open
329
square facing the Arena.
330
The dynamic evolution of odor isocontours gives a qualitative idea of
331
the odor impact expected from the emission, given the position and the
332
meteorological scenario. Yet, for a quantitative comparison we need more
333
synthetic descriptors which can be obtained from the statistical analysis of
334
the time series of odor concentration calculated for each grid point of the
335
computational domain.
336
Figure 6 shows the time series of odor concentration calculated during
337
wind event 1 for the grid point closest to the emission source S1 and for
338
point P1. According to the FIDOL methodology (see Environment Agency,
339
2011) the intensity and frequency of odor exposure are two of the main char-
340
acteristics necessary to assess the offensiveness of odors. Due to the short
341
averaging time and brief simulation period used in this work we can not use
342
the recommended regulation approach to assess odor impact. We propose to
15
343
use two odor impact criteria similar to those discussed by Griffiths (2014),
344
based on either intensity or frequency of odor impact events evaluated over
345
the time interval of interest (the 15 minute long period in our case). Specif-
346
ically, for the first odor criteria, we fix the frequency of exceedance (10%)
347
and derive odor concentration isocontours which can be compared against
348
threshold values; for the second odor criteria, we fix an odor concentration
349
threshold (Cref = 5 ou/m3 ) and derive maps of frequency of exceedance. Any
350
specific value of frequency of exceedance and odor concentration threshold
351
could be adopted to perform the kind of analysis we propose.
352
Figure 6 shows that near to the source (S1) the odor concentration is
353
quite large (653 ± 100 ou/m3 average value ± standard deviation, coefficient
354
of variation equal to 0.15) and only slightly changing over time; at point
355
P1, the odor intensity is significantly lower (11 ± 8.7 ou/m3 average value ±
356
standard deviation, coefficient of variation equal to 0.79) but the variability
357
in time is larger. The 90th percentiles are equal to 24.4 (indicated as dashed
358
thin line in the graph) and 802.6 ou/m3 (not shown) for P1 and S1; the
359
reference threshold concentration Cref (dashed thick line in the graph) is
360
exceeded 80% of time at P1 and 100% of time at S1.
361
Figure 7 shows the results of this analysis replicated for each point of
362
the computational grid: this can be used to compare and rank, according
363
to the two proposed assessment criteria, the odor impacts for Area 1 and
364
Area 2 for simulated wind events. Isocontours of 90th percentile of odor
365
concentration are shown in the top row and isocontours of the exceedance
366
frequency (C > Cref ) are shown in the bottom row. These maps show
367
the area in which any plotted concentration of odor is exceeded 10% of the
16
368
time at maximum, or the area in which detectable odor may be perceived
369
persistently (i.e. most frequently) in time.
370
Comparison between isocontours of 90th percentile calculated for Area 1
371
and Area 2 for wind event 1 (Figure 7, left half) indicates that the emission
372
will produce annoyance/severe annoyance at least 10% of the time along the
373
main street canyon in Area 1 and in front of the buildings facing the Arena
374
in Area 2. Detectable odor will be perceived for more than 50% of the time
375
in these areas.
376
The odor impact becomes even more significant for wind event 2 (Fig-
377
ure 7, right half). In this case, the emission will produce annoyance/severe
378
annoyance at least 10% of the time along the three street canyons for Area
379
1 and in a wide area close to the Arena in Area 2. Detactable odor will be
380
perceived for more than 50% of the time in even wider areas.
381
Figure 8 shows a final synthetic picture of odor impact given in the form
382
of odor roses, i.e. polar plots in which (i) the 90th percentile of odor concen-
383
tration (top half) or (ii) the percent frequency of exceedance of Cref (bot-
384
tom half) are plotted at reference distances (5, 25 and 45 m away from the
385
emission point) for each angular direction. Top and bottom rows in each half
386
represent the impact of the emission as is (untreated) or when the odor abate-
387
ment system is on. The radial scale of each plot is shown in the bottom right
388
corner. Consider first the impact of untreated source, S1 and S2, for wind
389
event 1 (first row, left half). The peak of odor concentration is found in the
390
south-west (S-W) direction, with odor concentrations as large as 20 ou/m3
391
25 and 45 m away from emission point S1 and as large as 40 ou/m3 25 and
392
45 m away from emission point S2. Minor peaks are also found along those
17
393
directions in which the wind and the local topography are “in phase”. The
394
frequency of exceedance of Cref (third row, left half) is up to 60% both 25
395
and 45 m away from emission point S1 in the S-W direction, and up to 55%
396
and 75% respectively 25 and 45 m away from emission point S2 in the same
397
direction. When the abatement system is on (second and fourth rows), the
398
odor impact becomes lower than 10 ou/m3 whichever the distance and an-
399
gular direction and Cref is exceeded 50% of the time at most. The right half
400
of Figure 8 shows the odor impact for wind event 2. In this case, the peak of
401
odor concentration is in the south-south-west (S-S-W) direction, with odor
402
concentrations as large as 40 and 76 ou/m3 respectively 5 m and 25 m away
403
from emission point S2. The frequency of exceedance is about 100% 25 m
404
and 45 m away from S2 in the S-W direction. These data indicate a more
405
intense and persistent odor impact for wind event 2. The odor impact can
406
be reduced in Area 1 treating the emission (second and fourth rows), with
407
annoying odor perceived less than 40% of the time 25 m away from the source
408
in the S-W direction. Annoying odor can be still perceived up to 60% of the
409
time 45 m away from the source in the W-S-W direction. The situation is
410
more critical for the source located in Area 2: even if the abatement system
411
reduces the odor impact, annoying odor will still be perceived 80% of the
412
time in the S-W direction 25 and 45 m away from the source.
413
4. Conclusions
414
In this work we propose the use of a fast response Eulerian-Lagrangian
415
type model to calculate the short term, short time average dispersion of odor
416
in urban areas. The model is based on a three dimensional computational 18
417
domain describing the urban topography at fine (one meter) spatial scale and
418
on highly time resolved (one minute frequency) meteorological data used as
419
inflow conditions.
420
We propose two odor impact criteria similar to those discussed by Griffiths
421
(2014) to assess odor impact: for the first odor criteria we fix the frequency
422
of exceedance (10%) to derive odor concentration isocontours which can be
423
compared against threshold values; for the second odor criteria we fix an
424
odor concentration threshold (Cref = 5 ou/m3 ) to derive maps of frequency
425
of exceedance. Simulations performed for the historical city of Verona for
426
two 15 minute long time periods show that the model can be used (i) to
427
comparatively evaluate and rank the odor impact of a given emission source
428
when located in alternative positions of the urban area; (ii) to check if end of
429
pipe technologies devised for odor control are effective or not to reduce the
430
impact.
431
We propose the odor rose plot of model output statistics (90th percentile
432
and exceedance frequency) as a simple graphical tool to compare odor impact
433
for different source locations and in different meteorological scenarios and to
434
evaluate the effectiveness of solutions proposed for odor impact mitigation.
435
Acknowledgements
436
The authors would like to thank Comune di Verona for geographical data
437
and R. Snidar and S. Rivilli for fruitful discussions. M.C. gratefully acknowl-
438
edges Michael Brown and R&D staff of Los Alamos National Laboratory for
439
the use of QUIC code and technical support, and MeteoVerona for the acqui-
440
sition of highly resolved wind data. N.P. gratefully acknowledges Laboratorio 19
441
di Olfattometria Dinamica, Udine, Italy for funding his PhD fellowship.
442
[1] Blanes-Vidal, V., Suh, H., Nadimi, E.S., Løfstrøm, P., Ellermann, T.,
443
Andersen, E.V., Schwartz, J., 2012, Residential exposure to outdoor
444
air pollution from livestock operations and perceived annoyance among
445
citizens, Environment International, 40: 44-50.
446
[2] Brattoli, M., de Gennaro, G., de Pinto, V., Loiotile, A.D., Lovascio, S.,
447
Penza, M., 2011, Odour Detection Methods: Olfactometry and Chemical
448
Sensors, Sensors, 11(5): 5290-5322.
449
[3] Campolo, M., Salvetti, M.V., Soldati, A., 2005, Mechanisms for mi-
450
croparticle dispersion in a jet in crossflow, AICHE J., 51(1): 28-43.
451
[4] Capelli, L., Sironi, S., Del Rosso, R., Centola, P., Rossi, A., Austeri, C.,
452
2011, Olfactometric approach for the evaluation of citizens’ exposure
453
to industrial emissions in the city of Terni, Italy, Science of the Total
454
Environment, 409: 595-603.
455
[5] DEFRA, Department for Environment, Food and Rural Affairs 2005,
456
Guidance on the Control of Odour and Noise from Commercial Kitchen
457
Exhaust Systems, Nobel House, 17 Smith Square, London SW1P 3JR.
458
[6] Dourado, H., Santos, J.M., Reis, N.C., Mavroidis, I., 2012, Numerical
459
modelling of odour dispersion around a cubical obstacle using large eddy
460
simulation, Water Science and Technology, 66(7): 1549-1557.
461
[7] Dourado, H., Santos, J.M., Reis, N.C., Mavroidis, I., 2014, Development
462
of a fluctuating plume model for odour dispersion around buildings,
463
Atmospheric Environment, 89: 148-157. 20
464
[8] Drew, G.H., Smith, R., Gerard, V., Burge, C., Lowe, M., Kinnersley, R.,
465
Sneath, R., Longhurst, P.J., 2007, Appropriateness of selecting different
466
averaging times for modelling chronic and acute exposure to environ-
467
mental odours, Atmospheric Environment, 41(13): 2870-2880.
468
469
[9] EN 13725:2003, Air quality − Determination of odour concentration by dynamic olfactometry.
470
[10] EN ISO 16911:2013, Stationary source emissions − Manual and auto-
471
matic determination of velocity and volume flow rate in ducts − Part 1:
472
Manual reference method.
473
[11] Environment Agency, 2011, Additional guidance for H4 Odor Manage-
474
ment: How to comply with your environmental permit, Environment
475
Agency Horizon House, Deanery Road Bristol, BS1 5AH.
476
477
[12] EPA 373/07 guidelines: Odour assessment using odour source modelling, 2007, www.epa.sa.gov.au/files/477172 guide odour.pdf
478
[13] Franzese, P., 2003, Lagrangian stochastic modeling of a fluctuating
479
plume in the convective boundary layer, Atmospheric Environment,
480
37(12): 1691-1701.
481
[14] Gailis, R.M., Hill, A., Yee, E., Hilderman, T., 2007, Extension of a
482
fluctuating plume model of tracer dispersion to a sheared boundary layer
483
and to a large array of obstacles, Boundary-Layer Meteorol., 122: 577-
484
607.
485
[15] Gowardhan, A.A., Pardyjak, E. R., Senocak, I., Browm, M.J., 2011, A 21
486
CFD-based wind solver for an urban fast response transport and disper-
487
sion model, Environmental Fluid Mechanics, 11(5): 439-464.
488
[16] Griffiths, K.D., 2014, Disentangling the frequency and intensity dimen-
489
sions of nuisance odour, and implications for jurisdictional odour impact
490
criteria, Atmospheric Environment, 90: 125-132.
491
[17] Hanna, S.R., Brown, M.J., Camell, F.E.; Chan, S.T., Coirier, W.J.,
492
Hansen, O.R., Huber, A.H., Kim, S., Reynolds, R.M., 2006, Detailed
493
simulations of atmospheric flow and dispersion in downtown Manhattan:
494
An application of five computational fluid dynamics models, Bulletin of
495
the American Meteorological Society, 87(12): 1713-1726.
496
[18] Huiling cui, Yao, R., Xu, X., Xin, C., Yang, J., 2011, A tracer experiment
497
study to evaluate the CALPUFF real time application in a near-field
498
complex terrain setting, Atmospheric Environment, 45: 7525-7532.
499
[19] Latos, M., Karageorgos, P., Kalogerakis, N., Lazaridis, M., 2011, Disper-
500
sion of Odorous Gaseous Compounds Emitted from Wastewater Treat-
501
ment Plants, Water Air Soil Pollut., 215: 667-677.
502
[20] Maizi, A., Dhaouadi, H., Bournot, P., Mhiri, H., 2010, CFD prediction of
503
odorous compound dispersion: Case study examining a full scale waste
504
water treatment plant, Biosystems Engineering, 106: 68-78.
505
[21] Manor, A., 2014, A Stochastic Single-Particle Lagrangian Model for
506
the Concentration Fluctuations in a Plume Dispersing Inside an Urban
507
Canopy, Boundary-Layer Meteorology, 150(2): 327-340.
22
508
[22] Michioka, T., Sato, A., Sada, K., 2013, Large-eddy simulation coupled to
509
mesoscale meteorological model for gas dispersion in an urban district,
510
Atmospheric Environment, 75: 153-162.
511
[23] Mussio, P., Gnyp, A.W., Hensha, P.F., 2001, A fluctuating plume dis-
512
persion model for the prediction of odour-impact frequencies from con-
513
tinuous stationary sources, Atmospheric Environment, 35: 2955-2962.
514
515
[24] Nicell, J.A., 2009, Assessment and regulation of odour impacts, Atmospheric Environment, 43 (1): 196-206.
516
[25] Nicolas, J., Cors, M., Romain, A.C., Delva, J., 2011, Identification of
517
odour sources in an industrial park from resident diaries statistics, At-
518
mospheric Environment, 44(13): 1623-1631.
519
[26] Olafsdottir, S., Gardarsson, S.M., 2013, Impacts of meteorological fac-
520
tors on hydrogen sulfide concentration downwind of geothermal power
521
plants, Atmospheric Environment, 77: 185-192.
522
[27] Piringer, M., Schauberger, G., Petz, E., Knauder, W., 2012, Comparison
523
of two peak-to-mean approaches for use in odour dispersion models,
524
Water Science and Technology, 66(7): 1498-1501.
525
[28] Pitton, E., Marchioli, C., Lavezzo, V., Soldati, A., Toschi, F., 2012,
526
Anisotropy in pair dispersion of inertial particles in turbulent channel
527
flow, Phys. Fluids, 24(7): 073305.
528
[29] Pullen, J., Vawda, Y., 2007, Review of Dispersion Modelling for Odour
529
Predictions, Science Report: SC030170/SR3 Environment Agency, Rio
530
House, Waterside Drive, Aztec West, Almondsbury, Bristol, BS32 4UD 23
531
[30] Schauberger, G., Piringer, M., Schmitzer, R., Kamp, M., Sowa, A.,
532
Koch, R., Eckhof, W., Grimm, E., Kypke, J., Hartung, E., 2012, Con-
533
cept to assess the human perception of odour by estimating short-time
534
peak concentrations from one-hour mean values. Reply to a comment
535
by Janicke et al., Atmospheric Environment, 54: 624-628.
536
[31] Sironi, S., Capelli, L., Centola, P., Del Rosso, R., Pierucci, S., 2010,
537
Odour impact assessment by means of dynamic olfactometry, dispersion
538
modelling and social participation, Atmospheric Environment, 44(3):
539
354-360.
540
[32] Sohn, J.H., Pioggia, G., Craig, I.P., Stuetz, R.M., Atzeni, M.G., 2009,
541
Identifying major contributing sources to odour annoyance using a non-
542
specific gas sensor array, Biosystems Engineering, 102(3): 305-312.
543
[33] Sommer-Quabach, E., Piringer, M., Petz, E., Schauberger, G., 2014,
544
Comparability of separation distances between odour sources and resi-
545
dential areas determined by various national odour impact criteria, At-
546
mospheric Environment, 95: 20-28.
547
[34] Zwack, L.M., Hanna, S.R., Spengler, J.D., Levy, J.I., 2011, Using ad-
548
vanced dispersion models and mobile monitoring to characterize spatial
549
patterns of ultrafine particles in an urban area, Atmospheric Environ-
550
ment, 45(28): 4822-4829.
24
(a)
(b)
N
U(Z)
S1 Z
P1
S1 P1
Area 1 Y X
(c)
S2 P2
Z S2 P2
Area 2 Y X
Figure 1: Aerial view of Verona (a) and areas selected for odor dispersion demo: (b) street canyons (Area 1) and (c) open square nearby the Arena (Area 2); potential positions of odor emission source are shown as (light gray) red circles (S1 and S2); points 50 m away from the source downstream the prevailing blowing wind direction (N-E) are shown as (dark gray) blue circles (P1 and P2).
25
(a)
(b)
(c) 12
Wind speed and approaching direction Event 1 Event 2
100
NE
W
2
SW
4
6
SE
8
E
Wind direction, [deg N]
NW
Event 1 Event 2
10 Wind speed, [m/s]
N
50
0
-50
8 6 4 2
Event 1 Event 2
S
0
2
4
0 6
8
10 12 14 16
Time, [min]
0
2
4
6
8
10 12 14 16
Time, [min]
Figure 2: Polar representation (a) and time series plots of wind direction (b) and wind speed (c) of wind data extracted for simulating odor dispersion: data are taken from meteorological station of Verona Golosine.
26
(a)
(b)
(c) 12
Wind speed and approaching direction
W
2
4
SW
6
8
E
100
50
0 Event 1 S1 P1 S2 P2
-50
SE S
0
(d)
2
4
0 0
SW
4
6
SE
8
2
4
E
6 8 10 12 14 16 Time, [min]
(f) Event 2 S1 P1 S2 P2
10
100 Wind speed, [m/s]
2
4
12
Event 2 S1 P1 NE S2 P2
Wind direction, [deg N]
Event 2
W
6
6 8 10 12 14 16 Time, [min]
(e)
NW
8
2
Wind speed and approaching direction N
Event 1 S1 P1 S2 P2
10 Wind speed, [m/s]
Event 1
NW
Wind direction, [deg N]
Event 1 S1 P1 NE S2 P2
N
50
0 Event 2 S1 P1 S2 P2
-50
S
0
2
4
8 6 4 2 0
6 8 10 12 14 16 Time, [min]
0
2
4
6 8 10 12 14 16 Time, [min]
Figure 3: Polar representation (a, d) and time series plots of wind direction (b, e) and wind speed (c, f) calculated in different points of the computational domain for wind events 1 (top row) and 2 (bottom row): lines with symbols correspond to (i) anemometric data used as inflow condition (10 m above ground) (red/green, solid), (ii) emission point position (1 m above roof level) (solid symbol, S1 blue/dark gray, S2 pale blue/light gray), (iii) control point position (1.5 m above ground) (empty symbol, P1 blue/dark gray, P2 pale blue/light gray).
27
Area 1 240 s
480 s
720 s ou/m
3
Wind event 2
Wind event 1
12 10 8 6 4 2 0
Figure 4: Isocontours of odor concentration calculated for Area 1 and wind event 1 and 2. Values are shown for a plane z = 1.5 m above the ground: snapshots are taken at every 240 s.
28
Area 2 480 s
720 s ou/m3 12 10 8 6 4 2 0
Wind event 2
Wind event 1
240 s
Figure 5: Isocontours of odor concentration calculated for Area 2 and wind event 1 and 2. Values are shown for a plane z = 1.5 m above the ground: snapshots are taken at every 240 s.
29
3
Odor concentration, [OU/m ]
1000
100
10
P1 S1
1
C90% Cref 0.1 0
100
200
300
400
500
600
700
800
900
Time, [s]
Figure 6: Time series of 30 seconds average odor concentration calculated at point P1 (closed symbol) and S1 (open symbol) for wind event 1: dashed lines represent 90th percentile of odor concentration for point P1 (thin dashed line) and a reference odor concentration threshold (5 ou/m3 , thick dashed line) sufficient to cause nuisance.
30
Wind event 1 Area 1
Wind event 2 Area 2
Area 1
Area 2
90th percentile
3
ou/m
12 10 8 6 4 2 0
Exceedances
31 % Exceed. 100 80 60 40 20 0
Figure 7: Statistics for odor impact assessment: (a) 90th percentile of odor concentration and (b) percent of exceedances (C > 5 ou/m3 ) during wind event 1 and 2 in Area 1 and 2.
90th percentile Wind event 1
Wind event 2
Untreated
R=45 m R=25 m R= 5 m
3
10 Ou/m
3
10 Ou/m
3
10 Ou/m
3
10 Ou/m
3
10 Ou/m
3
10 Ou/m
3
10 Ou/m
3
Treated
10 Ou/m
Area 1
Area 2
Area 1
Area 2
% Exceedances Wind event 1
Wind event 2
Untreated
R=45 m R=25 m R= 5 m
50%
50%
50%
50%
50%
50%
50%
Treated
50%
Area 1
Area 2
Area 1
Area 2
Figure 8: Odor rose of 90th percentile of odor concentration and % Exceedances for untreated and treated emission S1 and S2 and wind event 1 and 2.
32
Sample N.
T, [o C] RH, [%]
Q, [Nm3 /s]
C, [ou/m3]
U-1
31.3
24.1
5.4
5,000
U-2
29.5
22.4
6.2
3,800
U-3
32.2
28.7
5.9
5,000
Average
31.0
25.07
5.83
4,600
T-1
30.8
24.6
6.0
1,300
T-2
30.3
22.7
4.3
1,300
T-3
30.9
23.5
4.5
2,000
Average
30.7
23.6
4.93
1,533
Table 1: Results of odor source sampling: U (untreated) identifies odor emission with abatement system turned off, T (treated) identifies odor emission with abatement system turned on.
33